How are hospitals using 'AI that saves patients' lives'?
AI that can process a huge amount of data and find patterns is also useful in the medical field that collects a large amount of patient data, and in fact, in many hospitals, 'risk of readmission of patients who are about to be discharged' and 'patients develop sepsis'. AI is used to judge 'risk to do'. The Wall Street Journal, a major daily newspaper, reports on the use of AI in such hospitals.
How Hospitals Are Using AI to Save Lives --WSJ
Kaiser Permanente , an organization of the health maintenance organization that is an American medical insurance system and operates many medical institutions, has developed a prediction model called 'Advance Alert Monitor' that predicts the deterioration of the patient's condition. However, it is operated in the medical field. The Advanced Alert Monitor continuously scans patient data and assigns scores that predict the risk of being taken to the emergency room or dying, allowing doctors to prevent illness in advance. thing.
Dr. Vincent X. Liu, an intensive care expert at Kaiser Permanente, said, '(What the Advanced Alert Monitor does) is like looking for a needle in hay, the most of all patients. We have to screen people at high risk, 'he said, arguing that AI can be more efficient than human power to prevent patients from being too late.
The Advanced Alert Monitor is a system in which trained nurses remotely monitor the score instead of displaying it directly to the hospital staff to prevent the hospital staff from falling into 'alert fatigue'. increase. Only when the patient's score reaches a certain value will the remote nurse contact the ward nurse, where the patient will be formalized and the doctor will decide on a rescue program that includes transfer to the emergency room. is.
In a paper published in the New England Journal of Medicine in November 2021, the Advanced Alert Monitor was operated in 19 hospitals for about three years, compared to the case where the system was not operated. Improvements have been reported, including lower mortality rates, lower transfer rates to emergency rooms, and shorter hospital stays. At the time of writing, Kaiser Permanente operates advanced alert monitors at 21 hospitals, with nurses handling more than 16,000 alerts annually.
A research team led by Assistant Professor Cara O'Brien at Duke University Hospital found that commonly used models did not work well in their hospitals, so they used patient data collected at Duke University Hospital for septicemia. He said he decided to create his own machine learning model to predict the risk of. The research team trains algorithms based on data such as vital sign measurement, test results, and medication collected from more than 42,000 inpatients, and diagnoses the patient's risk of sepsis every 5 minutes. We have developed a machine learning model called.
The prediction results of the sepsis watch are displayed as a patient list color-coded into four by risk on the 'sepsis monitoring dashboard' that can be accessed on tablets and the like. During the 12-hour shift, one nurse will monitor the iPad dashboard and, depending on the situation, contact the emergency physician to discuss all patients at risk of sepsis. The doctor then independently reviews the medical records to determine if treatment for sepsis is needed.
Before the introduction of the sepsis watch, the proportion of sepsis patients treated appropriately at Duke University Hospital was about 31%, but after the introduction, the proportion of treatment increased to 64%. Mark Sendak, a clinical data scientist at Duke University Hospital, said mortality seems to be declining, even though final analysis is underway.
Not only Duke University Hospital is using AI to predict sepsis, but
In addition, HCA Healthcare data scientist Edmund Jackson and colleagues have applied the Spot platform to quickly show early signs of patient-threatening conditions such as traumatic shock, post-surgical complications, and patient illness. We have developed a broader program called 'Nate' to detect. Nate can also be applied to COVID-19, and it seems that it was possible to develop an algorithm that recommends the wearing of a ventilator to medical staff even during a pandemic.
HCA Healthcare's AI development team is working with clinical staff to determine which predictive models are useful and how to fit patient care. 'HCA Healthcare has a dedicated innovation team that goes to the hospital and works with healthcare professionals next to their beds,' said Michael Schlosser, senior vice president of care transformation and innovation at HCA. I don't suddenly appear and say, 'There is an AI trained to do XX for you.'
The Wall Street Journal also uses machine learning algorithms to identify high-risk patients from among those who have passed the deadline for colorectal cancer medical examinations, such as 'AI for diagnosing the risk of re-hospitalization of discharged patients'. We also touched on 'a system that identifies and recommends medical examinations.'
As AI is introduced in many hospitals, the challenge is 'when does AI not work and how can it be improved?' There are many discrepancies between the data used to build the algorithm and the actual data, and if defects are not detected properly, there is a risk of failure to diagnose critically ill patients or recommending harmful treatments. For example, at the University of Michigan, when a COVID-19 pandemic occurred, the AI sepsis diagnostic algorithm used up to that point could not properly distinguish between sepsis and COVID-19, so the algorithm was temporarily disabled. ..
'Currently, hospitals are overwhelmed by the number of AI models available,' said Karandeep Singh, assistant professor of the Medical AI Committee at the University of Michigan School of Medicine. He argued that he needed to understand when things didn't work as he did, and focus on problem-solving, not just availability.
Related Posts:
